Benchmark vectorized reverse and reverse_copy, use traits, optimize reverse_copy tail (#5493)

Co-authored-by: Stephan T. Lavavej <[email protected]>

Author: AlexGuteniev

benchmarks/CMakeLists.txt

@@ -118,6 +118,7 @@ add_benchmark(priority_queue_push_range src/priority_queue_push_range.cpp)
 add_benchmark(random_integer_generation src/random_integer_generation.cpp)
 add_benchmark(remove src/remove.cpp)
 add_benchmark(replace src/replace.cpp)
+add_benchmark(reverse src/reverse.cpp)
 add_benchmark(search src/search.cpp)
 add_benchmark(search_n src/search_n.cpp)
 add_benchmark(std_copy src/std_copy.cpp)

benchmarks/src/reverse.cpp

@@ -0,0 +1,53 @@
+// Copyright (c) Microsoft Corporation.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+#include <algorithm>
+#include <benchmark/benchmark.h>
+#include <cstdint>
+#include <vector>
+
+#include "skewed_allocator.hpp"
+#include "utility.hpp"
+
+template <class T>
+void r(benchmark::State& state) {
+    const auto size = static_cast<size_t>(state.range(0));
+    auto v          = random_vector<T, not_highly_aligned_allocator>(size);
+
+    for (auto _ : state) {
+        benchmark::DoNotOptimize(v);
+        std::reverse(v.begin(), v.end());
+    }
+}
+
+template <class T>
+void rc(benchmark::State& state) {
+    const auto size = static_cast<size_t>(state.range(0));
+    auto v          = random_vector<T, not_highly_aligned_allocator>(size);
+    std::vector<T, not_highly_aligned_allocator<T>> d(size);
+
+    for (auto _ : state) {
+        benchmark::DoNotOptimize(v);
+        std::reverse_copy(v.begin(), v.end(), d.begin());
+        benchmark::DoNotOptimize(d);
+    }
+}
+
+void common_args(auto bm) {
+    bm->Arg(3449);
+    // AVX tail tests
+    bm->Arg(63)->Arg(31)->Arg(15)->Arg(7);
+}
+
+
+BENCHMARK(r<std::uint8_t>)->Apply(common_args);
+BENCHMARK(r<std::uint16_t>)->Apply(common_args);
+BENCHMARK(r<std::uint32_t>)->Apply(common_args);
+BENCHMARK(r<std::uint64_t>)->Apply(common_args);
+
+BENCHMARK(rc<std::uint8_t>)->Apply(common_args);
+BENCHMARK(rc<std::uint16_t>)->Apply(common_args);
+BENCHMARK(rc<std::uint32_t>)->Apply(common_args);
+BENCHMARK(rc<std::uint64_t>)->Apply(common_args);
+
+BENCHMARK_MAIN();